for “Tsunami Warning System: Case Study of Intelligent Agents in...

CASE TEACHING NOTES for

“Tsunami Warning System: Case Study of Intelligent Agents in Action”

1

1 Introduction

This case study analyzes a tsunami warning system (TWS) from the perspective of an intelligent agent, the Tsunami Activity Reporter. Students receive a description of the reporter and an overview of a warning system from the National Oceanic and Atmospheric Administration and design an intelligent agent using the PEAS (performance measure, environment, actuators, and sensors) framework. It is targeted toward students in an upper division Introduction to Artificial Intelligence course. Careful choice of questions will allow the instructor to focus on critical thinking skills and adapt the case to non-technical audiences, such as K-12/CS0/CS1/CS2. With the addition of programming assignments the case study can form the basis for a term project in a knowledge-based systems course with implementations in a rule-based tool such as JESS (http://herzberg.ca.sandia.gov/) or CLIPS (http://clipsrules.sourceforge.net/). A systems engineering course can use the case study as context for analyzing sensor/software interfaces.

In this case, students are introduced to tsunamis and tsunami warning systems and asked to design an intelligent agent that will decide whether to issue a tsunami watch, warning, or information bulletin. Lives and property must be protected, but coastal residents will tend to ignore warnings if they “cry wolf” too often and alert people about a disaster than never comes. On the other hand, failing to alert residents of low-lying coastal areas can result in tragedy.

The March/April 2012 issue of IEEE Intelligent Systems [1] examines the close association of human, agents, and robots working in teams to accomplish complex tasks. A tsunami warning system is an example of such collaboration. A tsunami warning system collects data from seismometers and sea-level monitoring instrumentation attached to buoys in the world’s oceans, analyzes the data, assesses the potential impact of a tsunami, and disseminates appropriate warnings through various channels [2]. The Tsunami Warning System case study presents an opportunity to analyze the system from the perspective of an intelligent agent.

Various incarnations of this case study have been used in Introduction to Artificial Intelligence courses at both the undergraduate and graduate levels to frame discussions of agent architectures and the properties of the environments in which intelligent agents work. It has served as a homework assignment, in-class collaborative activity for small groups, and as an exam question. TWS reinforces concepts of agent architecture and critical thinking skills while presenting a real-world and familiar context for problem solving. It presents an opportunity to invite in experts from other disciplines and an opportunity to discuss the ethics involved in warning systems and the dangers of false negatives and false positives. Experience indicates that as students debate the capabilities of various architectures and apply their knowledge of environments, they gain a deeper understanding of issues involved in working with intelligent agents.

Case Teaching Notes for "Tsunami Warning System: Case Study of Intelligent Agents in Action "

2

2 Objectives

This case is designed to introduce students to agent architectures and environments in the context of a problem from earth science. They should have some familiarity with heuristics and problem solving methods.

This case can be presented during a 75-minute class meeting using the following timeline and the five questions included with the case:

• 5 minutes to organize into groups and hand out the case • 10 minutes to share homework solutions within the group • 35 minutes for group discussion • 20 minutes for reporting and discussing group results • 5 minutes for concluding remarks and presentation of the follow-on assignment.

If the class has a large number of groups, consider using the last few minutes of the group discussion time for group scribes to pair up and combine their reports, so that the subsequent number of groups reporting is cut by half.

Upon completion of Tsunami Warning System, students should be able to:

• Define and provide and example of an external performance measure • Identify the elements of an environment • Describe the properties of an environment in terms of whether it is

o accessible or inaccessible o deterministic or nondeterministic o episodic or nonepisodic o static or dynamic or semidynamic o discrete or continuous

• Identify actuators • Identify sensors • Identify goals and plans • Analyze and propose agent architecture

o Table lookup o Simple reflex o Goal-based o Utility-based

• Define and provide and example of an internal evaluation function, stating whether it is o Static o Dynamic

• Propose a utility function to select among multiple competing goals

3 Background

Russell and Norvig chapter 2 Intelligent Agents [3]

Tsunami warning system [2]


3

4 Preparation

Prior to approaching the case, introduce students to the fundamental task of artificial intelligence (AI), the concept of heuristics, the separation of data, operations, and control in an AI system, metaphor of a system as an agent, agent architectures, and the properties of environments in which the agents act. We want the students to understand the building blocks for agents and the issues associated with those blocks. It is helpful for students to have an introduction to concepts of goals, plans, and scripts and the notion that an intelligent agent can hold competing and conflicting goals. To interest them in the case study, spend a few minutes talking about or reviewing videos of recent tsunamis and discuss their impact and prevention.

5 Classroom Management

5.1 Pre-class Assignments

Pre-class assignments include completing the assigned reading, exploring related resources on the web, reflecting on the readings, and drafting a design of a solution by answering assigned questions.

5.1.1 Preparation Research

Students should explore the following references prior to using this case to become familiar with agent concepts and the domain. Provide the students the assigned readings and a list of resources one or two weeks prior to the meeting in which the case will be discussed and formally presented. One approach is to divide the background references among the students on 3- or 4-person teams so that each student will bring a unique perspective to the discussion and no one student needs to research all of the material. On the day the case is presented, the students should bring written responses to each of the assigned questions to share with their group.

Preparation appears as Assignments A-1 through A-4 in table 1. A brief description of the Tsumani Activity Reporter is shown in figure 1, with the assignment questions shown in figure 2. Figure 3 provides a more detailed description of the reporter, which is also included as attachment #1 of the assignment.


4

Table 1. Assignments for Tsunami Warning System.


5

Figure 1. Tsumani Activity Reporter.

Figure 2. The five Agent questions for this assignment.


6

Figure 3. An overview of the Tsunami Warning System.


7

5.1.2 Discussion Questions Prior to Starting Case

Students should prepare responses to each of the five Agent questions shown in figure 2, submit them to the instructor for review before class as stated in Assignment A-3 and have them available on the day they study the case as a group. The instructor can subsequently compare the draft responses to the group responses turned in as assignment A-6 to gain insight into how the students' concepts about agents and environments changed through the discussions.

5.2 Case Presentation

Begin class with a brief statement of the goals for the discussion (to examine agent architectures, environments, and implications). Distribute the case and related questions and instruct the groups to begin by identifying team members to act as time-keeper (to ensure that the team completes all tasks in the allotted time), discussion chair (to keep the conversation moving and cover all topics, as well as eliciting participation by all team members in the discussion), secretary (to record the discussion), and reporter (to share the results with the class following the group work). Let them know that at the end of the discussion each group will need to post its answers to the discussion questions and that each student must confirm that they have seen and agreed with the final version. Individual students might want to keep their own notes, as well, to use when completing subsequent homework assignments or study for exams. Each group should spend a few minutes reviewing the domain of the problem and sharing the results of their research into additional resources on tsunamis. Next, groups should take a few minutes to understand the differences in approach and assumptions they made before formulating the group’s response to the questions. Encourage students to reference the background readings in their responses where relevant, explain their design choices, and justify their descriptions of the properties of the environments. Questions you might to pose to groups as you observe the discussions:

• Is the change in water level at a specific single buoy significant by itself, or only in the context of changes at near-by buoys and/or over a period of time?

• Should such a system be fully automated, or should a human always remain in the loop before a tsunami warning is issued?

• Under what conditions would a table-driven agent be sufficient for dealing with this problem?

• What trade-offs would a utility-based agent need to make? • Can you come up with a formula for weighing multiple considerations? • What data and what knowledge are needed in order for this system to work? • How easy is it to represent the knowledge needed for this system to work?

5.2.1 Reporting Out

Twenty-five minutes before the end of class, allow each group to share their response to one or two questions (depending upon class size) and ask the other groups to critique the answers. Present questions such as the following to round out the discussion:

1. Do you think that a fully automated tsunami warning system is possible, given current technology? If not, would it be feasible in 5 years? Ever?


8

2. If such a system is developed, how can the ability to convince people that the warning or advisory is valid be incorporated into the system?

5.2.2 Conclusion/Closure

Just before class ends, tie together the points discussed by the students and remind them to post a polished version of the group response.

6 Follow-up Assignment Suggestion

Ask the students to apply what they have learned to a problem in a different domain, such as the Dynamic Disease Reporter as shown in figure 4.

7 Exam Questions Covered in the Course of the Case Study

1. Give examples of episodic and non-episodic environments and discussed what can be accomplished in one but not the other.

2. Consider an agent that plays soccer (or another team sport played on a field). Give an example of the reasoning such an agent might perform if it is architecture is based upon:

a. Table lookup b. Simple reflex c. Goal-based d. Utility-based


9

Figure 4. Dynamic Disease Reporter assignment.


10

REFERENCES and Resources

[1] Human-Agent-Robot Teamwork. IEEE Intellligent Systems. 27:2, 2012.

[2] The Tsunami Warning System: An international effort to save lives and protect property. http://www.ess.washington.edu/tsunami/general/warning/warning.html. Last accessed 6 FEB 2013.

[3] Russell, Stuart J. and Norvig, Peter. Intelligent Agents. In Artificial Intelligence: A Modern Approach. 3rd edition. Chapter 2. Prentice Hall, 2009.

for “Tsunami Warning System: Case Study of Intelligent Agents in...

Documents

Transcript of for “Tsunami Warning System: Case Study of Intelligent Agents in...